Molecular biology has historically described RNA as an intermediate between genetic information stored in DNA and protein synthesis. The estimated number of protein-coding genes is around 20,000 (Pertea et al., 2018). Classical approaches to classify RNAs with protein-coding potential—the messenger RNAs (mRNAs)—were typically based on the existence of open reading frame (ORF) longer than 300 nucleotides (nt), conservation, and/or functional domains (Dinger et al., 2008). Nevertheless, as protein-coding regions encompass only ∼2% of the human genome, the rest has been considered as “dark matter”. Detected RNAs not translated into proteins were named non-coding RNAs (ncRNA) and initially regarded as a transcriptional noise or the byproducts of genetic information flow from DNA to protein. Nevertheless, since the discovery of transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs), the number and understanding of new and putative functional ncRNAs have expanded. Moreover, the boundaries between the coding and non-coding RNAs have become more blurry. Evidence is emerging that some RNAs, initially classified as non-coding, hide small ORFs (sORFs, < 300 nt) encoding for small functional peptides- micropeptides. Currently, we know dozen of different ncRNAs, which can be can be classified as housekeeping or regulatory ncRNS, according to Szymanski et al. (2003).

Housekeeping ncRNAs are constitutively expressed and mostly well functionally characterized classes of rRNAs, tRNAs, small nuclear RNAs (snRNAs), small-nucleolar RNAs (snoRNAs), Ribonuclease P RNA (RNase P), Ribonuclease MRP RNA (MRP RNase, RNRP), and Telomerase RNA component (TERC). rRNAs are the most abundant class of RNAs in most cells, composing around 80% of cellular transcriptome. They serve as the essential binding site for ribosomal proteins within the assembled ribosome and contribute to the binding of extra-ribosomal factors and ribosome-associated proteins, resulting in the protein translation machinery (Noller et al., 2017; Simsek et al., 2017). tRNAs provide the interface between nucleic acids and proteins during translation by carrying an amino acid on its 3′ end and reading the mRNA by base-pairing induced by the ribosome, which uniquely determines the position of amino acids in proteins (Schimmel, 2018). snRNAs participate in the assembly and function of canonical spliceosomes (Wang and Burge, 2008). snoRNAs are localized to the nucleolus and guide the methylation and pseudouridylation of rRNAs, tRNAs, and snRNAs (Maxwell and Fournier, 1995). RNase P has a role in precursor-tRNA cleavage, RMRP in precursor-rRNA cleavage, and TERC in telomere synthesis (discussed later).

Regulatory ncRNAs are mostly produced in a cell- or tissue-specific fashion during certain stages of cell differentiation or organism development, or as a response to changes in the environment. They are still poorly understood and a very heterogeneous group that can act in different ways, from gene expression regulation to modulation of protein and RNA distribution within cells (Szymanski et al., 2003). They are divided based on their length into short (<200 nt) and long (>200 nt, lncRNAs) RNAs. Short ncRNAs consist of microRNAs (miRNAs), small interfering RNAs (siRNAs) and Piwi-associated RNAs (piRNAs). miRNAs are endogenous, single-stranded, 19-23 nt in length RNAs that can bind to a target mRNA with a complementary sequence to induce its cleavage, degradation, or interfere with translation. Similar in size, siRNAs are exogenous RNAs that undergo processing and function in post-transcriptional gene silencing (Carthew and Sontheimer, 2009). piRNAs are single stranded, 26-31 nucleotides long RNAs that form complexes with the piwi family of proteins. These complexes have a role in RNA and epigenetic silencing of transposons (Siomi et al., 2011). Longer than 200 nt, lncRNAs represent the most abundant, yet least understood class of RNAs, with an average length ~ 1000 nt (Ulitsky and Bartel, 2013). They share some features typical for mRNAs, such as transcription by the RNA-polymerase II (Pol II), 5′end cap, 3′end polyadenylation and presence of alternative splicing isoforms (Kopp and Mendell, 2018). However, compared to the mRNAs, they exhibit lower expression levels, more tissue-specific expression, and poor sequence conservation (Derrien et al., 2012; Djebali et al., 2012; Kopp and Mendell, 2018; Fazal et al., 2019). Although often considered as nucleus-enriched, lncRNAs exhibit variety of subcellular localization, which often helps to determine their biological function (Carlevaro-Fita and Johnson, 2019). Finally, circular RNAs (circRNAs) are a special class of RNAs with the 3′ and 5′ ends covalently linked, generally formed by alternative splicing of pre-mRNA (Salzman et al., 2012). They have been proposed to act as miRNAs sponges or even as templates for protein synthesis (Ragan et al., 2019).

Interest in the ncRNAs has been stimulated by the development of high-throughput OMICS technologies. Genome‐, transcriptome‐, translatome- and proteome‐wide measurements by the whole genome sequencing (WGS), RNA-sequencing (RNA-seq), ribosome profiling (Ribo-seq) and mass spectrometry (MS), respectively. In combination, these methods offer the possibility of a systematic analysis of different stages of gene expression (Ori et al., 2015; Wang et al., 2019). RNA-seq data have shown that up to 85% of the genome is transcribed and identified, among others, novel transcript isoforms, transcripts arising from intergenic regions, overlapping transcripts, and transcribed pseudogenes (Consortium, 2012; Djebali et al., 2012; Hangauer et al., 2013). Ribo-seq has shown widespread and pervasive translation on cytosolic RNAs, with surprisingly ~40% lncRNAs being engaged with the ribosome (Ingolia et al., 2009; Kearse and Wilusz, 2017). Reported ribosomal occupancy of RNAs indicated on the one side presence of different protein isoforms and regulatory upstream open reading frames ORFs (uORFs) from the mRNAs, and on the other, more exciting side, new ways of translational regulation and possible micropeptide production from lncRNAs (Morris and Geballe, 2000; Andrews and Rothnagel, 2014). It must be taken into account that the ribosomal occupancy of transcripts need not automatically lead to the production of stable, functional polypeptides, and that further evidence is needed in order to reclassify transcripts as indeed protein-coding (Guttman et al., 2013). MS has proven as a useful tool to inspect the postulated translational event, with developing proteogenomics approaches confirming the presence of some peptides encoded by previously non-coding regions (Slavoff et al., 2013; Fields et al., 2015; Wang et al., 2019). However, in order to omit the possibility of false-positive findings from MS, further functional studies on revealed peptides are needed, and these studies remain sparse.

The complexity of gene expression has in most cases been published on the levels of detection and its functional relevance remains elusive. Still, it has revealed that the distinguishment between mRNAs and ncRNAs is more challenging than initially assumed and that automatic gene annotation systems, although straightforward across large datasets, can sometimes be misleading. Traditional arbitrary ORF cutoff can lead to misclassification of some ncRNAs as mRNAs as they can by chance contain putative ORFs. This is especially true for lncRNAs, such as functionally characterized H19, Xist, Mirg, Gtl2, and KcnqOT1 (Prasanth and Spector, 2007). Some ncRNAs have evolved from the protein-coding genes, and so will keep certain features and homologies to mRNAs (Duret et al., 2006). For example, Xist has evolved into the ncRNA through the process of pseudogenization, during which proto-Xist had lost its protein-coding function and its flanking genes had turned into pseudogenes (Duret et al., 2006). On the contrary, micropeptide-encoding regions may be incorrectly classified as non-coding due to their size (Yeasmin et al., 2018). Next, the absence of ORF conservation does not guarantee an absence of protein-coding potential. Indeed, the majority of micropeptide-encoding regions are not conserved (Ji et al., 2015), suggesting their role in encoding evolutionary young proteins (Ruiz-Orera et al., 2014). Finally, some genes are bifunctional, and its products function independently both as RNAs and proteins. The first report of such a gene was the human Steroid Receptor Activator (SRA) (Lanz et al., 1999; Chooniedass-Kothari et al., 2004). SRA was initially characterized as ncRNA which co-activates steroid hormone receptors (Lanz et al., 1999) and later was revealed to also encode a functional protein (SRAP), which seems to modulate SRA activity (Chooniedass-Kothari et al., 2004).

Emerging discoveries in the ncRNA field have also raised the possibility that some ncRNAs affect mitochondrial biology. Mitochondria are crucial organelles for the integration of several key metabolic processes and the primary powerhouses in the cell (Spinelli and Haigis, 2018). The control of mitochondrial protein homeostasis is unique in that its components have origins in both mitochondria and nucleus ( Figure 1 ). Mitochondria contain their own circular genome (mtDNA). In humans, it is 16,569 bp in length and contains 37 genes- encoding for 2 rRNAs, 22 tRNAs, and 13 proteins of the oxidative phosphorylation (OXPHOS) system (Anderson et al., 1981) ( Figure 2 ). The rRNA coding sequences and all but one protein-coding sequences are separated by tRNAs and deprived of introns. The mtDNA is transcribed entirely from both strands, named heavy (H) or light (L). Transcription is initiated from the two H-strand (HSP1/2) and one L-strand promoter, located in the major non-coding region named “control region”, resulting in long polycistronic transcripts. LSP controls the transcription of eight tRNAs and the ND6 gene. HSP1 transcription produces a transcript containing tRNA^Phe, tRNA^Val, and the rRNAs, while transcription from HSP2 generates a transcript that spans almost the entire genome (Montoya et al., 1983; Chang and Clayton, 1984). The main proteins controlling the process are the RNA polymerase (POLRMT), two transcription factors (TFAM and TF2BM), transcription elongation factor (TEFM), and transcription termination factor (mTERF1) (Barshad et al., 2018). The “tRNA punctuation” model (Ojala et al., 1981) proposes that individual mRNA, rRNAs, and tRNAs are released from the polycistronic transcripts by the cleavage of tRNAs, which is in humans performed by endonucleases RNase P complex and ELAC2 (Holzmann et al., 2008; Brzezniak et al., 2011). After release, the rRNAs undergo chemical nucleotide modifications before becoming part of mitoribosome, the tRNAs undergo chemical nucleotide modifications, CCA addition at the 3′-end, deadenylation and finally aminoacylation, and the mRNAs get 3′ end polyadenylated (D’Souza and Minczuk, 2018). The half-life of mitochondrial transcripts and the decay of RNA intermediates are mediated by a complex of polynucleotide phosphorylase (PNPase) and SUV3 (Borowski et al., 2013). Finally, the mature mRNAs, tRNAs, and the assembled mitoribosome come together in the translation apparatus, for the synthesis of 13 subunits of OXPHOS system.

As mtDNA’s coding capacity is very limited, mitochondria are heavily dependent on the import of about 1,500 nuclear-encoded proteins. Besides, there have been indications that mitochondrial homeostasis is maintained not just through proteins, but also ncRNAs ( Figure 1 ). The presence of housekeeping mitochondrial nuclear-encoded ncRNAs has been postulated for decades. These ncRNAs include tRNAs (tRNA^Leu _UAA, tRNA^Gln _UUG, tRNA^Gln _CUG, tRNA^Lys _CUU), 5S rRNA, RMRP, and RNase P (Chang and Clayton, 1987a; Chang and Clayton, 1987b; Kiss et al., 1992; Yoshionari et al., 1994; Magalhaes et al., 1998; Puranam and Attardi, 2001; Holzmann et al., 2008). A systematic analysis of mitochondrial transcriptome further strengthened these claims. RNA-seq from 143B cells mitochondria and mitoplasts revealed the presence of several nuclear- and mitochondrial-encoded small RNAs and antisense transcripts (Mercer et al., 2011). Soon afterward, Rackham et al. (2011) observed by RNA-seq on HeLa cells that ncRNAs, excluding rRNAs and tRNAs, make up 15% of the human mitochondrial transcriptome, and identified three lncRNAs transcribed from the mtDNA. Follow-up studies have also reported the presence of ncRNAs encoded by the nuclear DNA, especially miRNAs and lncRNAs, within mitochondria across various cell types and tissues, suggesting that these ncRNAs may play important roles in the mitochondrial homeostasis (Kim et al., 2017b; Jeandard et al., 2019). The summary of the proposed nuclear-encoded ncRNAs is given in Table 1 .

Although detection of ncRNAs in mitochondria paved the way to more extensive research in this field with several examples of ncRNAs functionally described as directly impacting mitochondrial biology, these transcripts are far from being well characterized. It is important to mention that there are (still) many controversies and debates ongoing about the sole existence of ncRNA in mitochondria. The main obstacle presents the technical challenge of truly separating isolated and uncontaminated mitochondria from other membrane vesicles (endoplasmic reticulum (ER), the Golgi apparatus, the endosomes) they are tightly associated within the cell (Vendramin et al., 2017). Therefore, to assess the purity of mitochondria or mitoplasts, ER or other membrane vesicles should be used instead of cytosol or nucleus, which was not always the case. Mitoplasts—rather than mitochondria—should be subjected to RNase treatment before lysis in order to minimize the risk of contamination. Unfortunately, these control steps have not always been performed systematically, so the published data is to date a complicated topic of many debates (Vendramin et al., 2017). Moreover, implementation of high sensitive NGS techniques such as deep sequencing is likely to detect small amounts of contaminants, leading to data misinterpretation. Finally, as this field is still very fresh, many studies miss independent replicates and functional studies are published by one research group.

Despite these controversies, an increasing body of evidence has connected ncRNAs and their machinery with mitochondrial biology. In this review, we focus on classes of ncRNAs described to be functionally related with and/or localized in mitochondria: the housekeeping ncRNAs, miRNAs, and lncRNAs. We also take up the topic of mitochondrial micropeptides, recently discovered to be encoded within regions initially annotated as non-coding. Overall, we summarize knowledge on ncRNAs in mitochondrial biology and discuss their discovery, biosynthesis, import, and function in the context of both health and disease. Finally, we touch their potential as prognosis markers and therapeutic targets.

Several tRNAs, 5S rRNA, RMRP, and RNase P present housekeeping ncRNAs whose mitochondrial localization, transport, and function have been discussed for years. For some of them, their interacting RNA-binding proteins (RBPs) have been proposed and associated with mitochondrial import and function ( Figure 3 , Table 1 ). However, the exact import mechanism across mitochondrial membranes and the function of these ncRNAs remain unclear. It is important to note that reports of these ncRNAs have been sparse and therefore questionable, so more evidence is needed to confirm/deny their presence and role in mitochondria.

Nuclear-encoded tRNAs have been observed in mitochondria across many species, as most eukaryotes lack some of the essential tRNAs in their mtDNA and must import them (Tarassov et al., 2007; Schneider, 2011). Even though human mtDNA encodes all the necessary tRNAs, published data indicate that they are able to import some of the cytosolic tRNAs through conserved protein machinery. In vitro experiments have shown that the synthetic transcripts of yeast tRNAs could be internalized by the isolated human mitochondria (Kolesnikova et al., 2000; Entelis et al., 2001). Later, nuclear-encoded tRNAs have been detected in mitochondria (Rubio et al., 2008; Mercer et al., 2011), namely tRNA^Leu _UAA, tRNA^Gln _UUG, and tRNA^Gln _CUG. Gowher et al. (2013) successfully targeted yeast tRNA^Lys _CUU into human mitochondria in vivo, suggesting similarities in the tRNA import between the two species ( Figure 3A ). The current proposal by Gowher et al. (2013) is that tRNAs are recruited from the cytosol to the mitochondria with the precursor pre-KARS2 (mitochondrial lysyl-tRNA synthetase), helped by ENO2 (glyolitic enlolase). It is still unclear how the tRNA-pre-KARS2 complex then gets internalized into the mitochondrial matrix (Gowher et al., 2013; Kim et al., 2017b). Possible protein import pathway could consist of the translocase of the outer (TOM) and inner (TIM) mitochondrial membrane, as in yeast (Tarassov and Martin, 1996). Although the import of tRNA is yet to be fully understood, it could present a novel concept for therapy for disorders caused by defects in mtDNA-encoded tRNAs. Successful import of tRNA compensating the mutated mtDNA could rescue defects in mitochondrial translation. Rescue of mtDNA mutations by the import of designed tRNAs to mitochondria has been reported in vitro and in vivo (Salinas et al., 2008; Wang et al., 2012a), but more recent reports are missing.

Several studies have suggested that 5S rRNA is imported to the mammalian mitochondria (Yoshionari et al., 1994; Magalhaes et al., 1998). Entelis et al. (2001) suggested that mitochondrial 5S rRNA might substitute for its lost counterpart and be part of mitoribosome large subunit. Smirnov et al. (2008) proposed a model of mitochondrial 5S rRNA import ( Figure 3B ), starting with the recognition and transport of 5S rRNA from the nucleus to the cytoplasm by TFIIIA (Ciganda and Williams, 2011). In the cytosol, 5S rRNA was proposed to interact with pre-MRPL18 (precursor of mitochondrial ribosomal protein L18). This interaction might induce a conformational change in 5S rRNA that makes it recognized and bound by the mitochondrial enzyme Rhodanese, which helps it possibly translocate into mitochondria through a yet unknown mechanism. In the matrix 5S rRNA was proposed to associate with the mature MRPL18 and with mitoribosomes, affecting mitochondrial translation efficiency (Smirnov et al., 2010; Smirnov et al., 2011). However, as cryo-electron microscopy did not detect 5S rRNA within the mammalian mitoribosome 5S rRNA (Greber et al., 2015), its possible function in mitochondria remains enigmatic.

RMRP is a part of the RNase MRP, a ribonucleoprotein complex whose function has been discussed for decades. In the nucleus, it is involved in the pre-rRNA processing (Schmitt and Clayton, 1993; Chu et al., 1994; Goldfarb and Cech, 2017). In mitochondria, it was postulated to cleave RNA complementary to the light chain near the D-loop sites that mark the transition from RNA to DNA synthesis (Chang and Clayton, 1987b; Lee and Clayton, 1997). Three RNA-binding proteins (RBPs- HuR, PNPase, and GRSF1) have been implicated in the RMRP transport and role in mitochondria ( Figure 3C ). In the nucleus, RMRP is bound to HuR, which promotes its export to the cytosol in a CRM1-dependent manner (Noh et al., 2016). The exported RMRP might be then targeted into the mitochondrial intermembrane space through yet unknown mechanisms where PNPase was suggested to enable its import into the matrix (Wang et al., 2012b), after which its abundance in the matrix was reported to be increased through the interaction with GRSF1 (Noh et al., 2016). However, recent studies cast a shadow on the role of RMRP complex in mitochondria. Agaronyan et al. (2015) have shown that the RNA primer formation is a result of a premature arrest of the mitochondrial RNA polymerase after a G-quadruplex. Moreover, only the 3′ half (~130 nt) of RMRP could be found in mitochondria, indicating a processing that would result in a loss of catalytic activity (Esakova and Krasilnikov, 2010). These reports indicate that RMRP unlikely acts as an endonuclease in mitochondria. However, its interaction with GRSF1, an important component of the RNA granules (Antonicka et al., 2013; Jourdain et al., 2013), might still make it involved in the RNA metabolism.

RNase P processes the 5′ leader of precursor tRNA, which is a critical step of processing mitochondrial polycistronic transcripts (Ojala et al., 1981; Rackham et al., 2016). Two types of RNase P are known: ribonucleoproteins RNases P containing RNase P and protein-only RNases P (PRORP) (Lechner et al., 2015; Klemm et al., 2016). In the majority of species, including humans, it is assumed that the ribonucleoprotein RNase P acts in the nucleus and PRORP in mitochondria (Holzmann et al., 2008; Lechner et al., 2015). Strengthening this assumption, studies have reported that mammalian mitochondrial RNAse P does not require the catalytic RNA component for catalysis (Rossmanith et al., 1995; Holzmann et al., 2008). Nevertheless, RNase P was partially purified from HeLa cells mitochondria. Detected “mtRNase P”, together with the observed sensitivity of RNAse P to the nuclease treatment, suggested that RNAse P acts as a ribonucleoprotein also in mitochondria (Doersen et al., 1985). In addition, several groups indicated that mtRNase P is imported into the mitochondrial matrix through interaction with PNPase (Wang et al., 2010; Mercer et al., 2011; Noh et al., 2016) ( Figure 3D ). However, as so far functional RNase P ribonucleoprotein has not been reported in mitochondria, the existence of mtRNase P remains controversial (Jeandard et al., 2019).

hTERC is the RNA component of the human telomerase, where it serves as a sequence template for the telomere replication (Gall, 1990). As its sequence contains a region similar to an RMRP and RNase P short stem-loop that was proposed to enable their entry into mitochondria (Wang et al., 2010), hTERC was also proposed to be mitochondria-localized (Cheng et al., 2018). It was detected by the RT-PCR in purified mitoplasts, but as as a shorter, 195 nt-long transcript, which was termed TERC-53. Zheng et al. (2019) demonstrated that TERC-53 is mostly localized in the cytosol, where it regulates cellular senescence and is involved in cognition decline in mice hippocampus without affecting telomerase activity or mitochondrial functions. Having this in mind, the authors hypothesized that TERC-53 is exported from the mitochondria back to the cytosol (Cheng et al., 2018; Zheng et al., 2019). However, this hypothesis indicates hTERC processing occurring within the mitochondria, which has so far not been reported.

Vertebrate genomes contain thousands of miRNAs: according to MiRBase catalog, with the human genome containing 2,654 mature sequences (Kozomara et al., 2019). The biogenesis and biological functions of miRNAs have been widely studied in eukaryotic cells (Bartel, 2009) ( Figure 4 ). In short, miRNAs are transcribed from the intergenic regions or in antisense orientation to coding regions as the primary miRNA transcript (pri-miRNA). pri-miRNA is processed in the nucleus by Drosha and/or DiGeorge syndrome chromosomal region 8 (DGCR8). This results in premature miRNA (pre-miRNA) which is then bound by exportin 5 (XPO5). XPO5, along with RanGTP, enables the export of the pre-miRNA through the nuclear pore into the cytosol. There RNase Dicer (DICER1 in humans) cleaves it, producing mature double-stranded miRNA. From two strands, the “passenger strand” undergoes RNA degradation while the remaining “guide strand” associates with argonaute 2 (AGO2) and becomes part of a multiprotein RNA-induced silencing complex (RISC) (Han et al., 2006). The main function of miRNA within RISC is post-transcriptional gene regulation by promoting mRNA degradation or translational repression by sequence-specific binding to the target mRNA. mRNA degradation is achieved via AGO2 (Carthew and Sontheimer, 2009; Chekulaeva and Filipowicz, 2009). Translational control is mediated by GW182 (Czech and Hannon, 2011; Iwakawa and Tomari, 2015). Moreover, miRNAs have also been implicated in some non-canonical functions, such as direct transcription and chromatin state regulation in the nucleus, and even translational promotion (Vasudevan, 2012; Yao et al., 2019). Each miRNA can target multiple genes, enabling them to regulate the expression of over 60% of the human genes and therefore moderate any part of cellular biology (Bartel, 2009; Friedman et al., 2009). Focusing on mitochondria, based on their localization and genetic origin, three different classes of mitochondria-related miRNAs can be distinguished (1) cytoplasmic, nuclear-encoded miRNAs targeting mitochondria-related transcripts; (2) mitochondrial, nuclear-encoded miRNAs; and (3) mitochondrial, mtDNA-encoded miRNAs (Bandiera et al., 2013) ( Figure 4 ). The two latter classes, termed mitomiRs, are yet to be functionally deciphered.

The number of lncRNA genes in mammals varies broadly between different sources, from less than 20,000 to more than 100,000 in humans (Zhao et al., 2016; Kopp and Mendell, 2018). According to noncode.org, they are encompassing ∼144 000 loci in humans (Zhao et al., 2016). Intriguingly, although nucleus-enriched, lncRNAs have been observed in different cell compartments, including mitochondria (Dong et al., 2017). Their biological activities are highly influenced by their localization in the cell (Mercer and Mattick, 2013; Fatica and Bozzoni, 2014). lncRNAs have been suggested to regulate cellular biology via transcriptional regulation, organization of nuclear domains, and bindings to proteins or other RNAs (Ulitsky and Bartel, 2013; Kopp and Mendell, 2018). It is therefore not surprising that their disruption has been associated with different diseases (Briggs et al., 2015; Huarte, 2015; Uchida and Dimmeler, 2015).

lncRNAs can be functionally classified into those that act in cis, and those that act in trans (Kopp and Mendell, 2018). In cis , the lncRNA locus can regulate chromatin or gene expression of nearbye genes in at least three potential mechanisms: (1) DNA elements within the lncRNA promoter or locus carry the regulatory function, which is not related to the lncRNA or its production; (2) the act of transcription and/or splicing of the lncRNA affects nearby genes, irrespective of the transcribed lncRNA sequence; and (3) the lncRNA transcript alone affects the nearby genes, most commonly leading to the establishment of repressive or activating chromatin states. Some lncRNAs function in trans throughout the cell in, again, at least three potential mechanisms: (1) lncRNAs affect chromatin states and gene expression of distant genetic regions, (2) lncRNAs take part in the nuclear structure and organization (for example, as parts of speckles and paraspeckles), and (3) lncRNAs interact with proteins and/or other RNA molecules and modulate their expression and function (Lee, 2012; Rinn and Chang, 2012). Moreover, some transcripts initially annotated as lncRNAs are not non-coding, but actually encoding for biologically active micropeptides (Anderson et al., 2015; Matsumoto et al., 2017; Kopp and Mendell, 2018).

Over twenty lncRNAs have been described so far to affect the mitochondrial biology directly or indirectly. Some act in the cytosol, by regulating mitochondria-associated genes, often in interaction with miRNA, thus creating a complex mRNA-ncRNA regulation network. Other nuclear-encoded lncRNAs have been described to localize and act in mitochondria. As their transport mechanism into mitochondria is unknown their presence remains questionable. Finally, several lncRNAs have been discovered to be transcribed from mtDNA. These two latter mitochondria-localized, but origin-different lncRNAs could be refered to as nuclear-transported mitochondria-associated lncRNAs (ntmtlncRNAs) and mitochondria-encoded lncRNAs (mtlncRNAs) (Zhao et al., 2018).

Micropeptides are a class of small peptides encoded by a sORFs, without N-terminal signaling sequence and as such are released into cytoplasm immediately after translation. Due to their sORF that escapes automatic gene annotation, they tend to be overlooked and therefore misannotated as non-coding. Indeed, lncRNAs and TUFs (transcripts of unknown function) represent the greatest source for sORFs (Yeasmin et al., 2018). Although numerous ribosome profiling studies have reported substantial ribosome occupancy of the lncRNA transcripts, the MS and the proteogenomic approaches have confirmed only a small portion of them, numbers ranging from less than 100 to up to 1600 (van Heesch et al., 2019). With a lack of consensus in the datasets, the true coding potential of lncRNAs currently remains open to speculation. Several in-depth investigations have characterized lncRNA-derived micropeptides with important roles in the ion channel modulation (Anderson et al., 2015), cell signaling (Matsumoto et al., 2017) and RNA regulation (D’Lima et al., 2017). It is important to state that the mammalian mitochondrial proteome is surprisingly enriched in micropeptides, accounting for 5% of its proteins (Calvo et al., 2016). In recent years, several micropeptides within lncRNA were discovered and characterized with a role in mitochondria, some even encoded by the mtDNA (Kim et al., 2017a). Termed mitochondrial-derived peptides (MDPs) (Kim et al., 2017a), these mtDNA-encoded peptides- humanin, MOTS-c, and SHLPs were described as potential mitochondrial bioenergetics and metabolism regulators.

Mitoregulin (MOXI, MPM) has been discovered by four different groups recently as a muscle- and heart-enriched 56-amino acids inner mitochondrial membrane micropeptide encoded within LINC00116. It has a role in mitochondrial respiratory chain supercomplexes support, fatty acids oxidation, and Ca²⁺ dynamics (Makarewich et al., 2018; Stein et al., 2018; Chugunova et al., 2019; Lin et al., 2019). Lin et al. (2019) highlighted its importance in the muscle tissue, finding it upregulated during myogenic differentiation and knockout mice exhibiting smaller skeletal muscle fibers, worse muscle performance, and slower regeneration.

Humanin is a 24-amino acids micropeptide whose sORF is embeded within the 16S rRNA of mtDNA (Yen et al., 2013). It was initially discovered in the surviving cells of Alzheimer’s disease brain (Hashimoto et al., 2001), suggesting its neuroprotective and cytoprotective role that has later been investigated and acknowledged across various diseases (Hashimoto et al., 2001; Muzumdar et al., 2009; Bachar et al., 2010; Oh et al., 2011; Gong et al., 2014; Kim et al., 2018). It was shown to block apoptosis, improve insulin sensitivity, decrease inflammation, and reduce oxidative stress during aging (Guo et al., 2003; Muzumdar et al., 2009; Zhao et al., 2013; Sreekumar et al., 2016). Its effects are yet to be assessed for therapeutic purposes, especially in the treatments of diabetes and neurodegenerative disorders.

MOTS-c (mitochondrial open reading frame of the 12S rRNA type-c) is a 16-amino acids micropeptide with an sORF within the 12S rRNA mtDNA and reported to act in the cytoplasm (Lee et al., 2015). The micropeptide was found to target the methionine-folate cycle and de novo purine biosynthesis pathway, increase AICAR levels, and activate AMPK, by which it increases glucose utilization, fatty acid oxidation, and changes nucleotide metabolism. MOTS-c has been proposed as a biomarker for metabolic function, as it correlates with markers of insulin resistance and obesity (Du et al., 2018). In high fat diet-induced obese mice, it prevented obesity, fat accumulation, and hyperinsulinemia, making it a possible therapeutic target (Lee et al., 2015).

SHLPs (small humanin-like peptides) are a group of 6 peptides discovered by an in silico approach to be encoded in the 16S rRNA region of mtDNA in mice (Cobb et al., 2016). Each peptide is 20-38 amino acids long, and their names were given due to similar biological effects as Humanin. Each SHLP showed a unique expression pattern across different tissues. Incubation of each synthetic SHLP with cells affected cell viability, proliferation, and apoptosis differentially, suggesting a specific role of each. Moreover, SHLP2 and SHLP3 induced oxygen consumption rate (OCR) and increased cellular ATP levels, which indicated them as mitochondrial modulators (Cobb et al., 2016). Indeed, the administration of SHLP2 to a cellular model of macular degeneration rescued its defects in the OXPHOS and the mtDNA copy number, and induced anti-apoptotic effects, indicating its therapeutic potential (Nashine et al., 2018). In addition, an intracerebral infusion of SHLP2 increased glucose uptake and suppressed hepatic glucose production (Cobb et al., 2016). Further supporting their role as insulin sensitizers, both SHLPs promoted pre-adipocyte differentiation (Cobb et al., 2016). Similarly to humanin, the circulating levels of MOTS-c and SHLP2 declined with age, indicating that they are potential regulators of aging (Lee et al., Lee et al., 2015; Cobb et al., 2016).

Development of high-throughput OMICS techniques, especially the next-generation sequencing, has shed new light on the non-coding fraction of the genome. Transcription of the majority of the eukaryotic genome generates not only mRNAs but a much bigger fraction of different ncRNA species that show complex structure, patterns of expression and regulation. It is now becoming apparent that RNAs are not important for cell only in the context of mRNAs as intermediates between DNA and protein, but also as powerful players themselves by affecting basically any stage of gene expression. The now expanding RNA field highlights the importance of bioinformatics analysis in order to predict and examine existence, evolution, structure, and function of non-coding regions and transcripts. Focusing on mitochondria, dozens of ncRNAs acting in the cytosol have been described to indirectly influence mitochondrial biology, usually by targeting mitochondria-related, nuclear-encoded transcripts. More surprisingly, recent research indicated that the mitochondrial transcriptome could represent a mixture of the intrinsic transcriptome and complemented by some extrinsic RNA, implying RNA import ( Figure 1 ). Although dozens of papers reported ncRNAs in mitochondria, their existence is still under a question mark. Further research will need to identify their interacting partners and elucidate the molecular mechanisms behind their synthesis, transport, and function. Housekeeping ncRNAs have been proposed to have a mitochondrial localization even for decades, however, recent deeper insights into the mitochondrial biology have cast a shadow on their hypothesized role. It is clear that the re-evaluation of their presence and especially function in mitochondria is needed. Focusing on miRNA, they are well-described fine-modulators of gene regulation in the cytosol. It is not surprising that they can impact mitochondria by targeting its transcripts in the cytosol. Additionally, recent discoveries of mitomiRs suggest an attractive, even closer interplay of miRNAs and mitochondria occurring in mitochondria themselves. Yet, these findings are still a topic of many debates and therefore should be handled with caution. On the one side, the discovery of mitomiRs across different tissues and cell types by different techniques promises they are more than a false-positive finding. However, on the other side is the poor overlap between datasets that raises doubts concerning methods used. Focusing on lncRNAs, although they are among the least well-understood of these transcript species, they are slowly but surely emerging as important components of gene regulatory networks. Although the field of lncRNAs has just started to expand, published reports indicate that they influence mitochondria in different ways. Moreover, mtDNA seems to encode some lncRNAs itself. However, this field is still very fresh and further confirmation is needed, especially in the case of mitochondria-imported lncRNAs. Of clinical relevance, ncRNAs dysregulation has been noted in various mitochondria-related diseases, mostly cancer. Their association with tumorigenesis has been increasingly demonstrated. As ncRNAs often exhibit cancer-type-specific expression patterns (Iyer et al., 2015), targeting them could prove as a very selective and specific approach. Notably, they can be targeted by the antagomiRs or antisense oligonucleotides (ASOs) (reviewed by Matsui and Corey, 2017). Indeed, several pre-clinical studies have already demonstrated the therapeutic benefits of ncRNA inhibition. For example, inhibition of SAMMSON in melanoma xenografts suppressed the tumor growth (Leucci et al., 2016). ASOs targeting ASncmtRNA reduced the progression of renal adenocarcinoma and melanoma metastases in mice (Lobos-Gonzalez et al., 2016; Borgna et al., 2017). Finally, ncRNA-derived micropeptides, although biologically active as peptides, are especially interesting in terms of their discovery. As many ribosomal-profiling studies report significant ribosomal occupancy of non-coding transcripts, it is evident that further confirmation of these findings by mass spectrometry is needed in order to recognize the importance of these reported translational activities. Discoveries of mitochondrial-derived peptides and enrichment of the mammalian mitochondrial proteome in micropeptides suggest the organelle as an evolutionary playground for small proteins, either due to still unknown localization signals or import system or simply driven by the size or amino acid (positive charge) composition (van Heesch et al., 2019). This also promises that there could be many micropeptides hidden in the non-coding region, awaiting discovery and characterization. Of clinical interest, discovered mitochondria-derived micropeptides have exhibited a variety of cyto- and neuroprotective effects, and promising results of both in vitro and in vivo studies further strengthen their therapeutic potential. Overall, ncRNAs in mitochondria present a thought-provoking, but unfortunately still neglected field of study. It raises many interesting, but also challenging questions whose answers might be of clinical importance. It may reveal some enigmatic biological mechanisms (such as the RNA import in mitochondria) and eventually lead to the development of new therapeutic strategies for mitochondria-related diseases. However, before the field of ncRNA truly expands, there are still a lot of experimental approaches to be optimized and biological mechanisms to be deciphered to conclude their importance for mitochondria.

MG: conceived the topic for the review, wrote the manuscript and created the tables and figures. HP: helped shape the review, supervised the writing process, provided the critical feedback, contributed to the final version of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.